Bridging Large-Model Reasoning and Real-Time Control via Agentic Fast-Slow Planning

Dimensions overall score 7.0

• Agentic Fast-Slow Planning reduces lateral deviation by up to 45% compared to pure MPC and an A*-guided MPC baseline.result 90%
  Evidencepartial

  reducing lateral deviation by up to 45% and completion time by over 12% compared to pure MPC and an A*-guided MPC baseline.

  Implicationpartial

  Directly stated in abstract with specific numeric improvement

  Verificationpartial

  partial

• Agentic Fast-Slow Planning reduces completion time by over 12% compared to pure MPC and an A*-guided MPC baseline.result 90%
  Evidencepartial

  reducing lateral deviation by up to 45% and completion time by over 12% compared to pure MPC and an A*-guided MPC baseline.

  Implicationpartial

  Directly stated in abstract with specific numeric improvement

  Verificationpartial

  partial

• The Perception2Decision bridge compresses scenes into ego-centric topologies using an on-vehicle VLM detector to reduce bandwidth and delay while preserving interpretability.method 85%
  Evidencepartial

  Perception2Decision compresses scenes into ego-centric topologies using an on-vehicle Vision-Language Model (VLM) detector, then maps them to symbolic driving directives in the cloud with an LLM decision maker - reducing bandwidth and delay while preserving interpretability.

  Implicationpartial

  Directly described in abstract as a core component of the method

  Verificationpartial

  partial

• Semantic-Guided A* embeds language-derived soft costs into classical search to bias solutions toward feasible trajectories.technical 85%
  Evidencepartial

  Semantic-Guided A* embeds language-derived soft costs into classical search to bias solutions toward feasible trajectories

  Implicationpartial

  Directly described in abstract as a technical component of the method

  Verificationpartial

  partial

• Existing approaches that let LLMs generate trajectories directly are brittle, hard to verify, and latency-prone.limitation 80%
  Evidencepartial

  Existing approaches either (i) let Large Language Models (LLMs) generate trajectories directly - brittle, hard to verify, and latency-prone

  Implicationpartial

  Directly stated in abstract as limitation of existing approaches

  Verificationpartial

  partial

• Existing approaches that adjust MPC objectives online mix slow deliberation with fast control and blur interfaces.limitation 80%
  Evidencepartial

  or (ii) adjust Model Predictive Control (MPC) objectives online - mixing slow deliberation with fast control and blurring interfaces.

  Implicationpartial

  Directly stated in abstract as limitation of existing approaches

  Verificationpartial

  partial

• The Agentic Refinement Module adapts planner hyperparameters using feedback and memory.technical 85%
  Evidencepartial

  while an Agentic Refinement Module adapts planner hyperparameters using feedback and memory.

  Implicationpartial

  Directly described in abstract as a technical component of the method

  Verificationpartial

  partial

• The framework decouples perception, reasoning, planning, and control across natural timescales.method 85%
  Evidencepartial

  a hierarchical framework that decouples perception, reasoning, planning, and control across natural timescales.

  Implicationpartial

  Directly stated in abstract as a core design principle

  Verificationpartial

  partial

Competitive landscape

A hierarchical AI framework that bridges large-model reasoning with real-time control for autonomous driving, improving robustness and efficiency.